The purpose of the present study was to empirically investigate the stability and accuracy of one suggested method for matching test difficulty to examinee ability level. Students' answers to traditional classroom tests were rescored by computer as if the examinations had been flexilevel tests. The scores thus obtained were found to correlate highly with the traditional test scores (0.8994 to 0.9478), thereby indicating that flexilevel test scores are sufficiently stable and accurate to allow their use for classroom evaluation purposes. (Author)